When do you use component scanning versus explicit @Bean configuration?

I organize explicit configuration around capabilities such as messaging or payments rather than one miscellaneous configuration class. Library auto-configuration should be conditional and should back off for user-defined beans. In application code, I avoid declaring the same component through both scanning and a bean method because duplicate definitions and ambiguous injection are harder to diagnose than a deliberate registration strategy. A focused context test can assert that the capability contributes the intended beans and no accidental alternatives.